Digitizing our family's history in a day

2026-10-11

My father has a copy of a 129-page family-made book containing my great-grandmother’s life story plus decades of genealogy registers. I mentioned wanting to use AI to explore our family genealogy, he brought the book out, and so I photographed every page with my phone over a few minutes. 2-3 hours of effort later that same night, I’d produced a well-edited Markdown version for further AI analysis, plus a well-formatted PDF version complete with 57 restored photographs, and a validated .ged family tree file to complement it. Free-tier AI set up free-tier bulk OCR and formatted the content into markdown, then orchestrated the <$10 of image upscaling costs to restore and properly integrate the photographs into the page.

The steps were simple:

1. Photograph every page, which synced to OneDrive automatically and landed on my computer for further processing. Lighting was passable but not great - no problem.
2. Use OpenCode with the free Muse Spark 1.3 Contributor model to orchestrate the work. First, it walked me through setting up free capacity for Azure Document Intelligence for OCR, and then wrote a Python script to send each page through and assemble the transcript and manifest. It ran over 40 minutes, no babysitting required.
3. OpenCode/Muse did a proofreading pass, fixing punctuation and hyphenation, rejoining paragraphs across page breaks, standardizing headings, and recomposing photo pages as figures. Nothing reworded. Out comes a 12 MB letter-size PDF.
4. Muse cropped each on-page photo out of the overall iPhone photo images for AI upscaling. I’ll be honest, it didn’t do a great job here. I suspect Azure Document Intelligence or another product would have done this job very well but with only 57 photos, I simply screenshotted them off the image previews in the sequence they appeared, being sure I had good crops since I’d pay for the next step, upscaling.
5. Muse orchestrated bulk processing the photos through OpenAI’s GPT Image 2.5 Sunburst, their latest AI image model. It ensured vignettes were maintained while photos had a clean edge-to-edge or fade to pure white appearance so they integrated perfectly in the pure-white background of the digital PDF. Did some upscaling as well. Note: I have a version with original images too, since this is the first model that seems capable of faithfully maintaining details during upscaling, but it’s still a little suspect looking at family photographs with imagined details.
6. Muse transcribed pedigree charts from the document and cross-referenced from available online sources to produce a coherent genealogy in GEDCOM format (190 people), complete with clear references. Between the book’s notes and public records (weddings, births, deaths, newspaper articles, etc.) I was able to expand a maternal line no search engine had found.
7. Muse built a one-command local viewer and SVG export pipeline so I could easily render and share a family tree in our preferred formats.

The whole thing scales down to zero expertise with any of the bulk-processing tools, and scales up to any size.